Three streams of evidence, namely simulations with coupled models, feedback analysis in the tropical Pacific, and observation-based paleoclimate reconstructions, all support the expectation of a future mean El Niño-like temperature response to the positive radiative forcing resulting from a continued increase in atmospheric greenhouse gas concentrations.
 Natural and anthropogenic changes to the climate system alter the radiative balance and the system responds by adjusting its temperature, precipitation, wind and other quantities toward a new equilibrium consistent with the radiative perturbation. In other words, the “radiative forcing” associated with the changes stimulates an array of dynamical and thermodynamical feedback processes in the system which determine the nature of the temperature response and that of other climate parameters. Increasing greenhouse gas (GHG) concentrations in the atmosphere act to warm the system but there are uncertainties in both the magnitude and the pattern of the temperature response. Thus while global warming has been “detected” for the global mean, the local details of climate change remain ambiguous (IPCC2001, Chapter 12). For stronger forcing, such as that expected as the 21st century progresses or for some past climates, the climate change signal is expected to become much more prominent and identifiable.
 Forecasts of future climate change depend on modern climate models which are themselves approximate representations of the climate system. The evaluation of climate models (IPCC2001, Chapter 8) bases confidence on their ability to simulate the current climate, to reproduce the global mean temperature trend for the 20th century, and to reproduce features of past climates. The ability of models to simulate paleoclimates nominally offers direct information on their ability to simulate climate change in response to forcings other than those experienced during the recent historical period. However, paleoclimate reconstructions are themselves uncertain so that generally only the most robust features of the past climates are unambiguous. We nevertheless suggest that an “observed” feature of the climate of the last glacial maximum (LGM) can be linked via paleoclimate simulations to a prominent feature of future climate change.
2. Climate Change Models, Simulations, and Observations
 Two versions of the CCCma global coupled climate model (CGCM) are used to simulate current climate, 20th century warming, and potential future global warming as well as paleoclimatic global cooling. The common atmospheric component is a global spectral model with T32L10 resolution and a comprehensive range of physical parameterizations [McFarlane et al., 1992]. The atmospheric component is coupled to a full three-dimensional ocean models at a resolution of 1.8 × 1.8°L29. The first version of the coupled model and its climate is discussed in Flato et al.  and the second version, differing modestly in its ocean component in terms of ocean mixing and ice dynamics, in Flato and Boer .
 Long control climate integrations and a range of global warming simulations for the 20th and 21st centuries have been carried out with both models. The climate sensitivity (the equilibrium global mean temperature change for a doubling of CO2 with a mixed layer ocean version of the model) is 3.5°C in both cases. The transient climate response (the global mean temperature change at the time of CO2 doubling for a transient integration with CO2 increasing at 1% per year with a full ocean version of the model) is 1.96°C and 1.93°C respectively. The temperature response patterns are very similar as well, especially in the tropics. We concentrate here on near-equilibrium global warming simulations (1000 years of integration) with GHG and sulphate aerosols fixed at their year 2100 values based on the IPCC IS92a scenario (made with the first version of the model, see Boer et al. [2000a, 2000b]) and on a global cooling simulation representative of the LGM (over 900 model years of integration) with imposed ice sheet topography and a reduction of CO2 concentration from 330 to 235 ppm (made with the second version, see Kim et al. [2002, 2003]). Results are based on the average over the last 50 years of the simulations.
Figure 1 displays the near-equilibrium sea surface temperature (SST) changes simulated in these two cases. The similarity in spatial pattern is striking and, in particular, a mean El Niño-like warming pattern is a prominent feature of the global warming simulation and a mean La Niña-like pattern of the LGM simulation. Paleoclimate LGM SST reconstructions also show a La Niña-like pattern as seen in Figure 2 from CLIMAP  and CLIMAP , although the amplitude of the cooling is considerably larger in the model result as is discussed further below. Most global warming simulations made with other models also exhibit a mean El Niño-like warming [e.g., Meehl and Washington, 1996; Timmerman et al., 1999; Cai and Whetton, 2000] but there are exceptions with models simulating a bland tropical SST change pattern [Meehl et al., 2000] or even, in one case, a La Niña-like pattern [Noda et al., 1999]. Yu and Boer  show that for the CCCma model many of the same physical processes are invoked in the mean El Niño-like response as in the more usual transient El Niño behaviour.
 Simulations of the LGM with fully coupled models (i.e., with full three-dimensional dynamical ocean components rather than two-dimensional mixed layer ocean components or specified SSTs) are slowly becoming available. An early attempt by Bush and Philander  gives a clear La Niña-like response with a tropical Pacific cooling of 5 to 6 degrees, although the 15 year integration is far from equilibrium. Hewitt et al.  obtain a La Niña-like pattern in a 1000-year LGM simulation with the UKMO coupled model (HadCM3) although the tropical Pacific cooling is weaker than in the CCCma or Bush and Philander results. Shin et al.  obtain a weak La Niña- like pattern and a modest tropical Pacific cooling which is not unexpected given the NCAR model's comparative low climate sensitivity [IPCC, 2001, Table 9.1]. None of these simulations show the patches of warming seen in the CLIMAP reconstructions. The actual magnitude of tropical cooling at the LGM is a matter of continuing investigation. As discussed in Kim et al. , the CCCma tropical cooling is comparatively large as a consequence of the coupled model's climate sensitivity which is larger for cooling than for warming and for a full compared to a mixed layer ocean. The results nevertheless agree with many aspects of paleoclimate data and we argue below that the La Niña-like SST pattern of cooling in the tropical Pacific is the result of the coupled atmosphere/ocean dynamical feedbacks that are called into play for both climate warming and cooling and that this pattern is superimposed on the overall temperature change.
 A 270 year LGM integration reported by Kitoh and Murakami  obtains warming patches in the eastern tropical Pacific associated with weaker trade winds. This somewhat El Niño-like response to LGM conditions contrasts with the La Niña-like response in most other coupled models. However, that model, unlike most others, also exhibits a La Niña-like response, rather than an El Niño-like response, to GHG-induced warming as discussed in Noda et al. . Thus all models appear to be consistent in reversing the sign of the mean tropical Pacific temperature response pattern when the sign of the forcing is reversed. The majority of models, including the CCCma model, exhibit an El Niño-like response to positive forcing (warming) and a La Niña-like response to negative forcing (cooling) independent of the overall level of tropical warming (cooling).
 We appeal to the CLIMAP [1976, 1981] reconstructions of LGM sea-surface temperature change in Figure 2 as “observation-based” evidence of the pattern of temperature response to the negative forcing of the LGM. The simulated tropical LGM cooling is certainly larger than that of the CLIMAP reconstructions which are, however, themselves uncertain as to the overall magnitude of the SST change. More recent proxy estimates suggest a colder tropical Pacific than seen in CLIMAP. For example, snow line depression [Rind and Peteet, 1985], tropical ice cores [Thompson et al., 1995], Coral [Guilderson et al., 1994], pollen [Colinvaux et al., 1996], and ground water noble gas [Stute et al., 1995] all suggest a tropical cooling of more than 5°C which is in somewhat better agreement with the LGM simulation.
 There is a strong similarity in the patterns of SST change in the tropical Pacific in Figures 1 and 2 suggesting that the same physical processes determine them. The “climate sensitivity” is a measure of the strength of the feedbacks invoked in the system by the imposition of a radiative forcing. Boer and Yu [2003a] investigate the geographical distribution of feedbacks based on the vertically integrated energy budget written as dh′/dt = A′ + R′ = A′ + ΛT′ + f where dh′/dt is the change in storage of energy in the system, A′ the effect of changes in horizontal energy transports in the atmosphere and ocean, and R′ the change in the radiative flux at the top of the atmosphere. The radiative perturbation is the sum of the “radiative forcing” f and the “radiative feedback” expressed as a linear function of surface temperature in the form R′ = ΛT′ + f. The sign and strength of the feedback is measured by Λ. The radiative forcing arises from perturbations in the infrared (e.g., from changes in GHG concentration) and the solar (e.g., from changes in surface albedo, aerosols) components of the radiation stream. All terms, including the feedback parameter, are functions of location and time.
 If the climate system is to attain a new equilibrium for a given positive or negative forcing then for the global average
and the global feedback must be negative giving 〈T′〉 = 〈f′〉 = − 〈f〉/ where is the climate sensitivity parameter linking global mean temperature to global mean forcing. The “local contribution” to the global feedback parameter is obtained as
where = 〈Λl〉. The strength and geographical pattern of the feedback processes in the model measured in this way are displayed in Figure 3. Boer and Yu [2003a] discuss the implications of the geographical distribution of Λl and of the physical feedback processes represented. In particular, the local feedback processes of Figure 3 largely determine the temperature response pattern T′ which need not, and generally does not, resemble the forcing pattern f.
 Positive feedback acts to amplify the temperature response to the forcing and negative feedbacks to damp it away. For instance, high latitude regions of positive feedback result from changes in surface albedo associated with the retreat of ice and snow in a warmer climate and their advance in a colder climate. Over much of the rest of the globe, negative feedback results from changes in outgoing infrared radiation associated with changes in surface temperature, modulated by the changes in moisture and temperature distribution in the atmosphere. The region of positive feedback in the tropical Pacific stands out as the exception to the generally negative feedback in the tropics and is the result of both dynamic and thermodynamic feedbacks in the coupled system as discussed in Yu and Boer  and Boer and Yu [2003b]. The warming of the tropical Pacific associated with increasing GHGs expands the convection region over the warm western Pacific toward the east as SSTs warm sufficiently to permit convection. The result is an anomalous Walker circulation component with rising motion over the region into which the convection has expanded and sinking motion on both flanks which calls into play the Bjerknes  positive feedback mechanisms characteristic of El Niño-like warming. The same mechanisms apply in reverse in the case of a general cooling of the tropical Pacific. Precluding changes in ocean dynamics, by using a mixed layer rather than a full ocean component for instance, quells the positive feedback and hence the El Niño/La Niña-like responses seen in Figures 1 and 3.
 This positive feedback in the central tropical Pacific supports the El Niño-like response seen in the global warming simulations of the CCCma (and other) coupled models (Figure 1a) in the sense that the local temperature response to positive radiative forcing is amplified rather than damped away. It also supports the La Niña-like response to negative radiative forcing in the LGM case of Figure 1b. Thus, the patterns of temperature response in the tropical Pacific in both cases are a consequence of the same feedback processes operating in the coupled system.
 While we cannot observe the future, there is a considerable effort to observe the past as evidenced in the CLIMAP reconstructions of Figure 2 and other analyses which support a La Niña-like pattern of temperature change in the tropical Pacific. Thus the modelling results are internally consistent, reflect the feedback processes operating in the current climate system, and are supported by the results of observational-based paleoclimate reconstructions. The result is that the three streams of evidence, namely simulations with coupled models, the feedback analysis in the tropical Pacific, and the paleoclimate reconstructions, all support the expectation of a future mean El Niño-like temperature response to the positive radiative forcing resulting from a continued increase in GHG concentrations in the atmosphere. It is in this sense that we suggest that there is observational support for an El Niño-like pattern of future global warming.
 We greatly appreciate the work of Dave Ramsden, Cathy Reader, Warren Lee, and other members of CCCma in the production of the CGCM results and Steve Lambert for help with data issues. B. Yu has been supported by the Canadian Climate Variability (CLIVAR) Research Network.